Conventionally, for example, when a program is recommended to a user in television broadcasting, radio broadcasting and the like, a program matching information on preference of the user is selected on the basis of program information (program metadata) such an electronic program guide (EPG) or the like. A method for recommending a program to a user differs depending on a method of obtaining user taste data. The recommending method includes for example an initial interest registration method in which information on interests of a user is registered initially and a program is recommended on the basis of the information, a viewing history using method in which a history of programs viewed by a user in the past is used to recommend a program, or a collaborative filtering method in which a history of viewing by another user is used to recommend a program.
The initial interest registration method has the user register for example favorite program categories (for example dramas, variety shows and the like), favorite genres (detective stories, comedies and the like), or names of favorite talent at the time of a start of use, and obtains a name of a program to be recommended by performing matching between these pieces of information as keywords and program metadata.
The viewing history using method accumulates metadata of viewed programs each time the user views a program. When the history metadata is accumulated to a certain degree, the metadata is analyzed, and thereby information such for example as favorite program categories, favorite genres, or names of favorite talent is obtained. By performing matching between these pieces of information as keywords and program metadata, a name of a program to be recommended is obtained.
In a recording device using an HDD (Hard Disk Drive), for example, not only a history of viewing but also user operations such as recording programming, recording or the like may be accumulated as history information, and used to obtain taste information. In this case, it is possible to partly distinguish between programs that are viewed by the user without any particular interest and are in a state of being viewed because a television receiver, a radio or the like is on and programs viewed intentionally with more interest. Thus, information reflecting preference of the user more can be obtained.
The collaborative filtering method performs matching between a history of viewing and operations of a first user and histories of viewing of other users to search for a second user having a similar history of viewing to that of a first user, and obtains viewing or operation history data of the second user to extract and recommend programs that have been viewed by the second user but have not yet been viewed by the first user.
There is a technology that adds an n-dimensional attribute vector as program attribute information to a broadcasting program in advance, and can select a program to be recorded or a program to be reproduced by comparing a selection vector generated on the basis of an average value of each attribute item of attribute vectors of contents initially registered by the user and programs reproduced by the user or recorded by programmed recording and attribute vectors (for example Japanese Patent Laid-Open No. 2001-160955).
However, when a program is selected on the basis of the initial registration method, it is possible to reflect only fixed interests at the time of initial registration by the user. Besides, to obtain detailed information makes user registration operations complicated. On the other hand, when the number of pieces of information registered is reduced to simplify the operation of inputting registration information in initial setting, recommendation can be made only on the basis of rough information on preference of the user, thus resulting in lower accuracy in selecting programs suiting the preference of the user.
On the other hand, when a program is selected using a sum, an average or the like of metadata simply collected on the basis of a history of viewing of the user or the like as in the history using method or the like, programs accurately suiting the preference of the user cannot be recommended unless the history is accumulated to a certain extent. Further, in the history using method, relationship between metadata is obscured, and thus personalization cannot be made sufficiently. In addition, accumulating the history can cause an imbalance in weighting between items that tend to accumulate as a history, such for example as genre (items that tend to be detected as things that the user has a preference for) and items whose elements tend to spread as a history, such for example as starring (items that tend not to be detected as things that the user has a preference for).
Specifically, for example, in a case where a user is a fan of commentator A and thus likes to view “a live broadcast of a game played by baseball team B in which broadcast commentator A comments”, information “live broadcast of baseball” as a genre tends to accumulate as a history (information “live broadcast of baseball” is easily detected as a thing that the user has a preference for), whereas information “commentator A” does not tend to accumulate as a history (information “commentator A” is not easily detected as a thing that the user has a preference for). Hence, there occurs a case where a live broadcast of a game played by baseball team B in which broadcast another commentator comments is recommended, but a variety program on which commentator A appears is not recommended.
Also in the case where a program attribute vector is added to a broadcasting program in advance, and a program to be recorded or a program to be reproduced is selected by comparing a selection vector generated on the basis of an average value of each attribute item of contents initially registered by the user and attribute vectors of programs reproduced by the user or recorded by programmed recording and attribute vectors, as disclosed in Japanese Patent Laid-Open No. 2001-160955, a history of user operations is used, and therefore there may similarly occur an imbalance in weighting between items that tend to accumulate as a history and items whose elements tend to spread as a history such as starring and the like.
Further, for example, in a selection vector generated for a user who likes dramas and likes variety programs on which comedian A hardly appearing in dramas appears, and whose viewing ratio of variety programs to dramas is 2:8, starring B who frequently appears in dramas but is not an actor that the user particularly likes is accumulated as a history. Hence, a documentary in which starring B who frequently appears in dramas appears is recommended before a variety show in which comedian A appears.
In addition, when an important item for selecting programs differs (for example importance is attached to starring or importance is attached to contents) depending on the user, all items are calculated similarly, and thus preference unique to the user are not reflected in some cases.
Further, the collaborative filtering method uses mere information on preference of another user, and therefore makes it difficult to extract information indicating preference of each user in detail.